TITLE: Sample performance numbers for virtual and inline

[ This material comes from "C++ Strategies and Tactics" by
  Robert B. Murry, p. 131 ]

As an experiment, I built an abvious String class that used an
underlying char* representation. All the member functions were
small (one to five lines). I then built a program that sorted an
array of ten Strings. On three different platforms, I timed a
version that had every member function statically bound, but without
using any inlines; a version that had every member function except
the constructors dynamically bound (virtual); and a version that used
virtual functions and a trivial derivation from a virtual base class.
The optimizer was turned on in all cases. [The table below] has the
results, with all times normalized to the time it took for the version
that used the inline functions.


Platform/Compiler	Inline		No Inline	Virtual 	V. Base

Sparc10/USL C++3.0	   1		1.05		1.07 		1.10

SGI 4D/USL C++ 3.0	   1		1.06		1.08		1.13

IBM386X/Borland C++3.0	   1		1.10		1.13		1.21


Design for inheritance can have a significant effect on programs that
use the class heavily. [...] You should not pay this cost unless there
is some benefit. On the other hand, the performance impact is not so
great that you should warp your designs to avoid the use of virtual
functions and/or virtual base classes.